Acoustilytix

Acoustilytix automates detection, classification, and quantitative analysis of ultrasonic vocalizations (USVs) to support studies of emotional processing, brain neurochemistry, and neural function in animal models.

Key Features:

• Automated detection and classification: Uses machine learning methodologies to detect and classify USVs with recording-environment-agnostic performance, reducing manual scoring requirements.
• Sensitivity and precision metrics: Reports a sensitivity of 93% and precision of 73% for USV detection, compared with DeepSqueak's reported sensitivity of 88% and precision of 41%.
• Data export: Produces outputs of detected calls and associated parameters in CSV format for downstream analysis.
• Machine learning call-type prediction: Includes planned integration of machine learning-based call-type classification to recommend call types and refine accuracy as additional data are incorporated.
• Hand-scorer training module: Provides an automated training approach for human scorers with immediate corrective feedback based on expert annotations, improving inter-rater reliability (kappa 0.30–0.75) after training on 1,000–2,000 calls.

Scientific Applications:

• Neuroscience and behavioral research: Enables analysis of USVs to investigate emotional states and neurochemical processes in animal models.
• Cross-environment experiments: Supports analysis of USV recordings collected across diverse recording environments and experimental setups.
• Scorer standardization: Facilitates training and standardization of human annotators to improve inter-rater reliability in USV classification.

Methodology:

Employs machine learning algorithms trained on extensive datasets of USV recordings annotated by expert scorers, with a recording-environment-agnostic design for detection and classification.